Google has announced Gemini 4 Argon, its latest frontier AI model designed for complex, long-running tasks in software engineering, business work, and cybersecurity.
According to Google, trusted cyber defenders are currently testing Argon through its Fairwind Program. The company plans to expand access to developers, businesses, and consumers after further testing and safety improvements.
One of the main changes is its 1 million-token output limit, up from 64,000 tokens on the previous model. This allows Argon to handle longer tasks and produce much larger outputs in a single run.
Google says it already uses Argon internally for tasks such as debugging, research, code migrations, and algorithm development. The company also reported that Argon helped optimize a quantum computing algorithm and identify memory improvements that could free up more than 300 TiB of memory across its data centers.
For software development, Google said Argon achieved a 77.9% score on DeepSWE v1.1, a benchmark for real-world, long-term software engineering tasks. It is also being used to help migrate large C and C++ codebases to Rust.
Beyond coding, Google said Argon has been tested for finance, legal research, tax work, and other enterprise tasks. It also supports visual tasks such as analyzing charts, understanding long videos, and working with information across multiple documents.
Cybersecurity is another focus. Google said Argon can identify, validate, and patch software vulnerabilities. The company is initially providing trusted cybersecurity defenders with access to the model without some of its usual cyber-related restrictions.
Google is also working on safeguards against misuse, prompt injection attacks, and other risks before wider availability. The company said Argon will initially become available to paid API customers and Google AI Ultra subscribers.
Google is currently testing Argon and plans to release it to developers, businesses, and consumers after further testing.